Object Counter (YOLO_V3)

Object counter is a toolkit that uses YOLO V3(you only look once version 3) algorithm. It makes an object detection on images/videos and count the number of objects present in the image/video.

this video test the toolkit on part of video of captain marvel trailer and here some examples of prediction on images

this toolkit allow you to make predictions using pre-trained weights or train the network in your own dataset for making predictions. if you have to count a lot of objects daily like goods you can use this toolkit to count for your in less than second.

Requirements

• Python 3.*

• Numpy

• OpenCV

• Pytorch

  there are three cases to run this toolkit:

• make predictions on images

• make predictions on videos

• train the model on custom dataset

  Setup For Predictions

  After download and extract the repository

  Weights

  Add your weights in weights directory, or download coco weights for yolov3 of tiny yolo by running these commands yolo -> $ sh weights/download_weights.sh tiny yolo -> $ sh weights/download_weights_tiny.sh

Configurations

Add your configuration file for your model in config directory, or use yolov3 of tiny yolo configurations models files

Images

images.py is the script that is responsible for managing predictions on images

usage: images.py [-h] [--images IMAGES] [--outputs OUTPUTS] [--bs BS]
                 [--confidence CONFIDENCE] [--nms_thresh NMS_THRESH]
                 [--cfg CFGFILE] [--weights WEIGHTSFILE] [--reso RESO]

YOLO v3 detection Module

optional arguments:
  -h, --help            show this help message and exit
  --images IMAGES       Image / Directory containing images to perform
                        detection upon
  --outputs OUTPUTS     Image / Directory to store detections
  --bs BS               Batch size
  --confidence CONFIDENCE
                        Object Confidence to filter predictions
  --nms_thresh NMS_THRESH
                        NMS Threshhold
  --cfg CFGFILE         Config file
  --weights WEIGHTSFILE
                        weightsfile
  --reso RESO           Input resolution of the network. Increase to increase
                        accuracy. Decrease to increase speed

Videos

video.py is the script that is responsible for managing predictions on videos

usage: video.py [-h] [--bs BS] [--confidence CONFIDENCE]
                [--nms_thresh NMS_THRESH] [--config CONFIGFILE]
                [--weights WEIGHTSFILE] [--outputs OUTPUTS] [--reso RESO]
                [--video VIDEOFILE] [--cam CAMERA]

YOLO v3 Detection Module

optional arguments:
  -h, --help            show this help message and exit
  --confidence CONFIDENCE
                        Object Confidence to filter predictions
  --nms_thresh NMS_THRESH
                        NMS Threshhold
  --config CONFIGFILE   Config file
  --weights WEIGHTSFILE
                        weightsfile
  --outputs OUTPUTS     Image / Directory to store detections
  --reso RESO           Input resolution of the network. Increase to increase
                        accuracy. Decrease to increase speed
  --video VIDEOFILE     Video file to run detection on
  --cam CAMERA          use camera to make detections

Setup For Training

After downloading and extracting the repository COCO DataSet For downloading the COCO dataset run this command $ sh data/get_coco_dataset.sh Custom DataSet Get into data directory and put our dataset in this directory (to manage the files) the dataset structure must be like coco data set

• images: should be in directory named images
• labels : should be in directory named labels, the label of each image must has same name but the extinction be .txt
• data configuration file should be in config directory and be like this

  classes=80
  train=data/coco/trainvalno5k.txt
  valid=data/coco/5k.txt
  names=data/coco.names
  backup=backup/
  eval=coco

Train

train.py is the script that is responsible for managing the training process

usage: train.py [-h] [--epochs EPOCHS] [--image_folder IMAGE_FOLDER]
                [--batch_size BATCH_SIZE]
                [--model_config_path MODEL_CONFIG_PATH]
                [--data_config_path DATA_CONFIG_PATH]
                [--weights_path WEIGHTS_PATH] [--class_path CLASS_PATH]
                [--conf_thres CONF_THRES] [--nms_thres NMS_THRES]
                [--n_cpu N_CPU] [--img_size IMG_SIZE]
                [--checkpoint_interval CHECKPOINT_INTERVAL]
                [--checkpoint_dir CHECKPOINT_DIR] [--use_cuda USE_CUDA]
                [--weights WEIGHTSFILE]

optional arguments:
  -h, --help            show this help message and exit
  --epochs EPOCHS       number of epochs
  --image_folder IMAGE_FOLDER
                        path to dataset
  --batch_size BATCH_SIZE
                        size of each image batch
  --model_config_path MODEL_CONFIG_PATH
                        path to model config file
  --data_config_path DATA_CONFIG_PATH
                        path to data config file
  --weights_path WEIGHTS_PATH
                        path to weights file
  --class_path CLASS_PATH
                        path to class label file
  --conf_thres CONF_THRES
                        object confidence threshold
  --nms_thres NMS_THRES
                        iou thresshold for non-maximum suppression
  --n_cpu N_CPU         number of cpu threads to use during batch generation
  --img_size IMG_SIZE   size of each image dimension
  --checkpoint_interval CHECKPOINT_INTERVAL
                        interval between saving model weights
  --checkpoint_dir CHECKPOINT_DIR
                        directory where model checkpoints are saved
  --use_cuda USE_CUDA   whether to use cuda if available

Credits

• YOLO v3 tutorial from scratch
• PyTorch YOLOv3